Just some example workflows of how to use vace to control your generations.
works on 6gb vram. And sometimes control fail which is to be expected.
Has some simple workflows for video control with depth/cannY7openpose
Video extension and spline control
If you don't want to use the prompt caching replace it with standard clip and prompt encode nodes.
Also made a video that probably badly explains things
Think this was where i got the model from:
https://huggingface.co/lym00/Wan2.1-T2V-1.3B-Self-Forcing-VACE-Addon-Experiment/tree/main
Description
This is mostly a mess.
You'd probably want to learn from the other one.
FAQ
Comments (13)
It might be just me, but I find the workflow a bit convoluted and unclear. What's the purpose of the 2nd start image? Is it for style/element replacement? Is it mislabeled and supposed to be end image? Why is there spline control AND reference video, but reference video is strictly required? The note nodes seem random and references in them are not clear and contain many typos. Lots of nodes are bypassed without explanation. It generally seems like a quick draft rather than something ready for release. I haven't used VACE before, only standalone spline control, so I might be missing something obvious.
My suggestion would be to
1. redo the note nodes and clear up what they are actually referring to
2. reorganize workflow by using groups and fast group bypasser node to make it more obvious what is optional and what is not for a given flow.
3. update civitai description to at least describe what the workflow is supposed to do.
Sorry if this comes across as rude, it's certainly not meant to be. I'm just majorly confused by this workflow.
Well either things are simple, but then they don't do much or complicated and do a lot. So i went with a in approach with some notes here and there that hopefully helps out.
Forgot the load nodes even had names on them. so the second is meant for the end or whatever else.
The reference video isn't even plugged in. but you have to option to plug it into the control instead.
Since i tend to move thing around and make everything into spaghetti and plug thing to new ones whenever i want something new i guess I'm more used to messing with workflows. You don't want to see my main image workflow lol
As it's plugged in by default it's just the start image and the spline that's connected. run the workflow once with the output disabled to load the spline then set your movements.
The two nodes that are bypassed have a note describing their purpose
It's indeed for you to try to experiment since it's not really feasible to make something for every scenario you might want specially without making it unsurmountable for beginners.
I might make some simpler dedicated workflows to accompany it
Feel free to leave any more questions
@JustSomeGuy Thanks for your reply! Here's what I have tried with your workflow:
1. upload starting image
2. adjust resize node values
3. start and abort workflow to refresh spline editor
4. draw spline path
5. replace your backend nodes with clip text encode nodes and clip loader
6. unmute video combine node
7. bypassed video loader node
8. ran the workflow
I can't get anything other than purple noise with the spline indicator superimposed on it.
In the state you published the workflow it does not produce an intuitive output. Feel free to check for yourself!
Some of the spaghetti might need connecting, but I struggle to find the intended path.
I'd appreciate your help.
@xnapx Well I did check it myself when i made it and it worked then. Check again later when I've updated and maybe that'll help clear things up
The updated version should be more understandable
@JustSomeGuy I only saw this just now. Thanks a lot for splitting up the individual workflows. Much easier to debug now. And a big thanks for taking the time to make the video! Got it all working. (And as a side effect, I'll be looking into implementing a backend solution for my own setup, so thanks for bringing that to my attention, too!)
i want to try it, i downloaded and i get size error , i tried downloading your vids, no metadata...
i added the images, changed the video contro, and im getting this in the console:
Prompt executed in 0.30 seconds
got prompt
Prompt executed in 0.31 seconds
got prompt
Prompt executed in 0.36 seconds
got prompt
Prompt executed in 0.37 seconds
got prompt
Prompt executed in 0.43 seconds
doesnt do anything, doesnt error either..
Well there are several possibilities. either you've forgotten to unmute the output nodes. or maybe you've not used or replaced the nodes for prompt caching.
Later on today I'll be intending on making some improved explanations. So if you don't manage to fix it you can look again here when I've updated
@JustSomeGuy yeah output was muted, i didnt know that will stop the whole workflow from executing xD
idk how to hide the red box.. if i set border 0 it doesnt seem to affect the output.
@JustSomeGuy "Loading aborted due to error reloading workflow data"
Hopefully the update is more understandable
@2P2 Not sure what that means. maybe you didn't give it any inputs? anyhow try the update instead